1. What is the root cause of prompt injection?
2. What distinguishes indirect prompt injection from direct prompt injection?
3. Which of these could carry an indirect prompt injection? Select all that apply.
4. Why is prompt injection unlikely to be solved the way SQL injection was?
5. Which change most reduces the severity of a successful prompt injection?
6. Why does upgrading to a more capable model not solve prompt injection?
7. An assistant answers questions from an internal wiki that any employee can edit. What is the principal risk?
8. What is the correct way to treat text produced by a language model before using it elsewhere?
9. Which of these are reasons agentic systems raise the stakes? Select all that apply.
10. What is the most important design principle for an approval gate?